Method

Segmentation-Aware 3D Object Detection [SA_3D]
[Anonymous Submission]

Submitted on 1 Feb. 2019 04:55 by
[Anonymous Submission]

Running time:0.3 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
Given the huge success recently in 2D object
recognition, it is natural to obtain these 2D
priors before moving into3D. In this work, we
specifically leverage the success of instance
segmentation and classification from the Mask-
RCNN network. By segmenting the desired object
instances out in 2D and back-project them into
3D, we remove the background noise and only have
a limited number of 3D points to process. A final
3D box estimation module is applied on each group
of 3D segmented object with the prior of its
class.
Parameters:
\learningrate=0.001
Latex Bibtex:

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Car (Detection) 86.36 % 66.69 % 55.18 %
Car (3D Detection) 41.70 % 31.21 % 25.96 %
Car (Bird's Eye View) 58.98 % 47.00 % 39.47 %
Pedestrian (Detection) 68.55 % 57.85 % 50.45 %
Pedestrian (3D Detection) 39.20 % 32.58 % 27.65 %
Pedestrian (Bird's Eye View) 47.45 % 35.93 % 34.99 %
Cyclist (Detection) 22.79 % 18.51 % 15.33 %
Cyclist (3D Detection) 17.99 % 14.15 % 13.52 %
Cyclist (Bird's Eye View) 26.45 % 20.56 % 16.42 %
This table as LaTeX


2D object detection results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot



2D object detection results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot



2D object detection results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker